How to compile binary with libtorch without CUDA

1

Is that possible somehow? I looked at the Caffe2 and Torch cmake files and there did not seem to be a straightforward way to do it (but I’m a newbie with C++ compilation and cmake).

2

I am not sure what you want. But the NO_CUDA=1 flag will disable all cuda stuff.

1 Like
3

I want to have no dependencies on CUDA when using libtorch to compile binaries, as I’m not going to be using a GPU.

I’m now thinking the way to do this is to compile pytorch without CUDA, is that possible? There’s a year old issue where the answer is no it isn’t.

4

As I said above, running the compilation with NO_CUDA=1 env variable set will do a compilation and install without any cuda dependency.

1 Like
5

Sorry, and thanks.

How could one compile it without installing it? I don’t want it to overwrite my existing installation, which I assume will happen when doing python3 setup.py install.

6

You can do python3 build to only build cpp dependencies.

1 Like
7

How do we set the No_CUDA = 1 flag? In the Cmake file or something else?

8

I set the environment variable NO_CUDA as 1 in my windows OS. It is not working. CMake still asked me cuda stuff.
The error message is following

– Selecting Windows SDK version 10.0.17763.0 to target Windows 10.0.18363.
CUDA_TOOLKIT_ROOT_DIR not found or specified
– Could NOT find CUDA (missing: CUDA_TOOLKIT_ROOT_DIR CUDA_NVCC_EXECUTABLE CUDA_INCLUDE_DIRS CUDA_CUDART_LIBRARY)
CMake Warning at C:/libtorch/share/cmake/Caffe2/public/cuda.cmake:31 (message):
Caffe2: CUDA cannot be found. Depending on whether you are building Caffe2
or a Caffe2 dependent library, the next warning / error will give you more
info.
Call Stack (most recent call first):
C:/libtorch/share/cmake/Caffe2/Caffe2Config.cmake:88 (include)
C:/libtorch/share/cmake/Torch/TorchConfig.cmake:40 (find_package)
CMakeLists.txt:4 (find_package)

9

reply to myself,
just use the libtorch that does not need cuda.
release: https://download.pytorch.org/libtorch/cpu/libtorch-win-shared-with-deps-1.5.0.zip

debug: https://download.pytorch.org/libtorch/cpu/libtorch-win-shared-with-deps-debug-1.5.0.zip

10

Updated answer:

# Bash
USE_CUDA=0 python setup.py install
# CMD (Windows)
set USE_CUDA=0
python setup.py install
# Powershell
$env:USE_CUDA="0"
python setup.py install